Data to 7 Oct 2026

Unpacking Failure Modes of Generative Policies: Runtime Monitoring of Consistency and Progress

arXiv 2410.04640 ↗First posted 6 Oct 2024cs.RO cs.AI cs.LGcited by 63

On the map: Offline RL and imitation learning from data → Language-conditioned policies and VLAs

Abstract (arXiv)

Robot behavior policies trained via imitation learning are prone to failure under conditions that deviate from their training data. Thus, algorithms that monitor learned policies at test time and provide early warnings of failure are necessary to facilitate scalable deployment. We propose Sentinel, a runtime monitoring framework that splits the detection of failures into two complementary categories: 1) Erratic failures, which we detect using statistical measures of temporal action consistency, and 2) task progression failures, where we use Vision Language Models (VLMs) to detect when the policy confidently and consistently takes actions that do not solve the task. Our approach has two key strengths. First, because learned policies exhibit diverse failure modes, combining complementary detectors leads to significantly higher accuracy at failure detection. Second, using a statistical temporal action consistency measure ensures that we quickly detect when multimodal, generative policies exhibit erratic behavior at negligible computational cost. In contrast, we only use VLMs to detect failure modes that are less time-sensitive. We demonstrate our approach in the context of diffusion policies trained on robotic mobile manipulation domains in both simulation and the real world. By unifying temporal consistency detection and VLM runtime monitoring, Sentinel detects 18% more failures than using either of the two detectors alone and significantly outperforms baselines, thus highlighting the importance of assigning specialized detectors to complementary categories of failure. Qualitative results are made available at https://sites.google.com/stanford.edu/sentinel.

Key facts. Cited by 63 papers in the map (data through 7 Oct 2026). Placed in Language-conditioned policies and VLAs, one of 13 directions in Offline RL and imitation learning from data. Builds on 74 of its 96 references that are in the map. First posted on arXiv 6 Oct 2024.

Builds on

74 of 96 references.

Closest ancestors · 1

Same direction, newest first

2023Model-Based Runtime Monitoring with Interactive Imitation Learning26 Oct

Other references · 73

2017Attention Is All You Needcited by 59,932
2015Deep Residual Learning for Image Recognitioncited by 51,809
2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
2020Denoising Diffusion Probabilistic Modelscited by 20,851
2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
2018Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actorcited by 4,635
2015Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learningcited by 4,416
2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,384
2021On the Opportunities and Risks of Foundation Modelscited by 3,961
2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,888

38 more

2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,175
2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,730
2015VQA: Visual Question Answeringcited by 2,489
2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
2023Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardwarecited by 1,490
2018A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attackscited by 1,472
2024Octo: An Open-Source Generalist Robot Policycited by 1,427
2021Calibrate Before Use: Improving Few-Shot Performance of Language Modelscited by 1,371
2022Inner Monologue: Embodied Reasoning through Planning with Language Modelscited by 1,057
2024Grounded SAM: Assembling Open-World Models for Diverse Visual Taskscited by 914
2024DROID: A Large-Scale In-The-Wild Robot Manipulation Datasetcited by 801
2021Generalized Out-of-Distribution Detection: A Surveycited by 796
2021A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantificationcited by 777
2021What Matters in Learning from Offline Human Demonstrations for Robot Manipulationcited by 699
2024SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilitiescited by 679
2020AWAC: Accelerating Online Reinforcement Learning with Offline Datasetscited by 647
2023BridgeData V2: A Dataset for Robot Learning at Scalecited by 634
2020A Unifying Review of Deep and Shallow Anomaly Detectioncited by 423
2020Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awarenesscited by 389
2022VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Trainingcited by 384
2024Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperationcited by 377
2024Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robotscited by 373
2016Kernel Mean Embedding of Distributions: A Review and Beyondcited by 360
2023Vision-Language Foundation Models as Effective Robot Imitatorscited by 344
2023Robots That Ask For Help: Uncertainty Alignment for Large Language Model Plannerscited by 298
2023Reconstructing Hands in 3D with Transformerscited by 298
2017Deep Imitation Learning for Complex Manipulation Tasks from Virtual Reality Teleoperationcited by 286
2018HG-DAgger: Interactive Imitation Learning with Human Expertscited by 245
2023RVT: Robotic View Transformer for 3D Object Manipulationcited by 224
2023Tracking Anything with Decoupled Video Segmentationcited by 197
2024PIVOT: Iterative Visual Prompting Elicits Actionable Knowledge for VLMscited by 187
2023REFLECT: Summarizing Robot Experiences for Failure Explanation and Correctioncited by 171
2023Physically Grounded Vision-Language Models for Robotic Manipulationcited by 163
2023Vision-Language Models as Success Detectorscited by 138
2019Self-Supervised Correspondence in Visuomotor Policy Learningcited by 106
2020Human-in-the-Loop Imitation Learning using Remote Teleoperationcited by 103
2022Denoising diffusion models for out-of-distribution detectioncited by 102

50 most cited of 73 All 73 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 2; 2025: 18; 2026: 43.

63 papers.

Sort:

By area:

2025SAFE: Multitask Failure Detection for Vision-Language-Action Modelscited by 74
2025RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Modelscited by 56
2025From Foresight to Forethought: VLM-In-the-Loop Policy Steering via Latent Alignmentcited by 51
2026Robometer: Scaling General-Purpose Robotic Reward Models via Trajectory Comparisonscited by 51
2025Can We Detect Failures Without Failure Data? Uncertainty-Aware Runtime Failure Detection for Imitation Learning Policiescited by 49
2025Failure Prediction at Runtime for Generative Robot Policiescited by 36
2025Uncertainty-aware Latent Safety Filters for Avoiding Out-of-Distribution Failurescited by 30
2025Do What You Say: Steering Vision-Language-Action Models via Runtime Reasoning-Action Alignment Verificationcited by 25
2025CUPID: Curating Data your Robot Loves with Influence Functionscited by 21
2026Scaling Verification Can Be More Effective than Scaling Policy Learning for Vision-Language-Action Alignmentcited by 15
2026CycleVLA: Proactive Self-Correcting Vision-Language-Action Models via Subtask Backtracking and Minimum Bayes Risk Decodingcited by 13
2025Real-Time Out-of-Distribution Failure Prevention via Multi-Modal Reasoningcited by 10

38 more

2025Mobi-π: Mobilizing Your Robot Learning Policycited by 9
2025Guardian: Detecting Robotic Planning and Execution Errors with Vision-Language Modelscited by 8
2026RC-NF: Robot-Conditioned Normalizing Flow for Real-Time Anomaly Detection in Robotic Manipulationcited by 7
2026Rewind-IL: Online Failure Detection and State Respawning for Imitation Learningcited by 7
2025Uncertainty Comes for Free: Human-in-the-Loop Policies with Diffusion Modelscited by 6
2026Failure Identification in Imitation Learning Via Statistical and Semantic Filteringcited by 6
2026Hide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoringcited by 6
2024The Duality of Generative AI and Reinforcement Learning in Robotics: A Reviewcited by 5
2024RoboMD: Uncovering Robot Vulnerabilities through Semantic Potential Fieldscited by 5
2025From Demonstrations to Safe Deployment: Path-Consistent Safety Filtering for Diffusion Policiescited by 4
2026Uncertainty Quantification for Flow-Based Vision-Language-Action Modelscited by 4
2026HarnessWAM: Bridging Prediction and Deliberation in World Action Modelscited by 4
2025KDPE: A Kernel Density Estimation Strategy for Diffusion Policy Trajectory Selectioncited by 3
2026Tri-Info: Generalizable, Interpretable Failure Prediction for VLA Models via Information Theorycited by 3
2025Hierarchical DLO Routing with Reinforcement Learning and In-Context Vision-language Modelscited by 2
2025EVE: A Generator-Verifier System for Generative Policiescited by 2
2026AEGIS: A Backup Reflex for Physical AIcited by 2
2026ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policiescited by 2
2026PATCH: Action-Chunk-Conditioned Latent Patch Innovation Monitoring for Robot Manipulationcited by 2
2026VLA-FAIL: Efficient Task Failure Detection for Finetuned Vision-Language-Action Modelscited by 2
2026Learning Robust Execution in Robotic Manipulation with Agentic Reinforcement Learningcited by 2
2026SAFECAST: Robust Failure Detection for VLA Policies with Contrast-Set Training and Calibrationcited by 2
2026AutoIntervene: Calibrated Intervention for Action-Chunking Imitation Learning Policiescited by 2
2025Improving the Resilience of Quadrotors in Underground Environments by Combining Learning-based and Safety Controllerscited by 1
2026Describe-Then-Act: Proactive Agent Steering via Distilled Language-Action World Modelscited by 1
2026CMP: Robust Whole-Body Tracking for Loco-Manipulation via Competence Manifold Projectioncited by 1
2026VLAConf: Calibrated Task-Success Confidence for Vision-Language-Action Modelscited by 1
2026GUARD: Grounding Uncertainty and Ablation-Based Risk Detection for Diffusion-Based VLAscited by 1
2025Enhancing Video-Based Robot Failure Detection Using Task Knowledgecited by 0
2026RAPID: Reconfigurable, Adaptive Platform for Iterative Designcited by 0
2026How VLAs Fail Differently: Black-Box Action Monitoring Reveals Architecture-Specific Failure Signaturescited by 0
2026Grasp-Then-Plan with Failure Attribution: A Closed Two-Stage Framework for Precise and Generalizable Robotic Manipulationcited by 0
2026Sensitivity Shaping for Latent Modelingcited by 0
2026Perturbation-Based Epistemic Uncertainty for Failure Detection in Vision-Language-Action Modelscited by 0
2026SynCity 3000: Bootstrapping Scene-Scale 3D Diffusioncited by 0
2026FailBench: How Reliable are VLMs at Judging Robot Task Success?cited by 0
2026FailureSpot: Label-Efficient Timestamp-Level Failure Detection for Vision-Language-Action Modelscited by 0
2026No Free Checker: A Survey of Verifiers for Robot Policiescited by 0

50 most cited of 63 All 63 (JSON)

Similar, not cited

Close in wording, no citation either way